Beyond Brainstorming: What Drives High-Quality Scientific Ideas? Lessons from Multi-Agent Collaboration

While AI agents show potential in scientific ideation, most existing frameworks rely on single-agent refinement, limiting creativity due to bounded knowledge and perspective. Inspired by real-world research dynamics, this paper investigates whether structured multi-agent discussions can surpass solitary ideation. We propose a cooperative multi-agent framework for generating research proposals and systematically compare configurations including group size, leaderled versus leaderless structures, and team compositions varying in interdisciplinarity and seniority. To assess idea quality, we employ a comprehensive protocol with agent-based scoring and human review across dimensions such as novelty, strategic vision, and integration depth. Our results show that multi-agent discussions substantially outperform solitary baselines. A designated leader acts as a catalyst, transforming discussion into more integrated and visionary proposals. Notably, we find that cognitive diversity is a primary driver of quality, yet expertise is a non-negotiable prerequisite, as teams lacking a foundation of senior knowledge fail to surpass even a single competent agent. These findings offer actionable insights for designing collaborative AI ideation systems and shed light on how team structure influences creative outcomes.

To this end, we define quality as a composite of five dimensions: novelty, feasibility, impact, coherence, and ethical soundness

Cognitive Stimulation and Process Constraints

A central principle in group creativity is cognitive stimulation: exposure to others’ ideas can activate novel associative pathways, enabling insights that may not emerge in isolation (Paulus and Nijstad 2003; Paulus 2000; Nijstad and Stroebe 2006). This is the primary rationale behind brainstorming. However, collaboration introduces process losses. Production blocking, where participants must wait their turn to contribute, disrupts thought processes and leads to idea loss (Paulus and Nijstad 2003). Evaluation apprehension, or fear of negative judgment, can inhibit the sharing of unconventional ideas (Nijstad and Stroebe 2006). These effects motivate a core research question: Can a collaborative system overcome its inherent process losses to provide net benefits over solitary ideation? Optimizing idea generation thus requires balancing stimulation with structure, motivating our exploration of ideal group configurations and interaction protocols (Coskun et al. 2000).

Implications for Multi-Agent Simulation of Ideation

Together, these theories do not prescribe a single ideal structure for collaboration. Instead, they define a multidimensional design space. Our work operationalizes these theoretical insights in a controllable multi-agent framework that simulates scientific ideation. By systematically varying parameters such as group size, communication structure, and agent composition, we aim to uncover design principles that support high-quality idea generation. This framework enables us to map the landscape of collaborative creativity and inform the development of more effective ideation systems: whether human-led, AI-driven, or jointly orchestrated.